1 / 24

Grid Operations SA1 Status Report

Grid Operations SA1 Status Report. Steven Newhouse (substituting for Maite Barroso Lopez) EGEE-III Final Review, 23-24 June, 2010. SA1 Activity Overview. 28 countries, 175 FTE. Grid Operations.

candid
Télécharger la présentation

Grid Operations SA1 Status Report

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Grid OperationsSA1 Status Report Steven Newhouse (substituting for Maite Barroso Lopez) EGEE-III Final Review, 23-24 June, 2010

  2. SA1 Activity Overview 28 countries, 175 FTE SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  3. Grid Operations Operate the EGEE production infrastructure, providing a high quality service to all the application groups Enabled through support structures including: • Regional Operation Centres • Global Grid User Support (GGUS) • Release and deployment management • Grid security coordination • Operational procedures and tools Structural changes needed in order to transition to a sustainable model SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  4. Size of the infrastructure Number of EGEE-III certified sites (April 2010): 317 (in 52 countries) Computing resources, doubled since last year: • 243,020 cores (139,000 last year) • 500 MSI2k (212 MSI2k last year) (MSI2k = Normalised CPU time to a reference value of Mega SPECint 2000) Storage resources: • 40 PB disk, 61 PB tape (25 PB disk, 38PB tape last year) SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  5. Usage of the infrastructure (I) CPU utilization over time: Consistent doubling every 12-18 months Usage from non-LHC communities stable LHC usage growing, real activity: Real data processing, simulations, analysis SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  6. Usage of the infrastructure (II) Number of jobs • Average of 15 million jobs per month (14M last year) • 525.000 jobs/day (480.000 jobs/day last year) • Non-LHC VOs with stable workload • LHC running increasingly high workloads • anticipate millions / day soon SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  7. Site availability and reliability (I) No changes in algorithm regarding last year • Service available if any of the service instances are available (logical OR) • Site available if all services are available (logical AND of all service types) EGEE League Tables (monthly reports) published and followed up with ROCs • Also available for top 10 VOs Underperforming sites requested to report justifications, collected in a wiki page • https://twiki.cern.ch/twiki/bin/view/EGEE/MonthlyAvailability SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  8. EGEE League Tables April 2010 Most regions have a good performance well above targets (70% availability and 75% reliability) April had highest average availability in the last year: 91% New ROCs started with lower A/R levels than experienced ones, took a few months to achieve targets SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  9. Site availability and reliability (II) Sites with availability <50% for three consecutive months are removed from the Production infrastructure • 7 sites removed, from AsiaPacific and Russia Simplification of downtime declaration rules • To enforce scheduled downtimes • Unscheduled downtimes has a big impact in site availability! And to users! • Ratio of scheduled/unscheduled downtimes doubled between May-09 and Apr-10 • Sites are doing a better job following the procedures and pro-actively informing their user communities of planned maintenances Most regions have a good performance well above targets • 70% availability • 75% reliability SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  10. Site availability seen by VOs Views from WLCG dashboards Built on top of VO specific grid monitoring probe results Using same grid monitoring framework SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  11. Site availability seen by VOs Views from the Complex Systems VO Monitoring Built on top of VO specific grid monitoring probe results Using same grid monitoring framework SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  12. Release and deployment management Move to Staged rollout as transition to EGI model • Controlled deployment of middleware updates into production • Get early feedback on the release under a production load • Done by a few sites on behalf of the community (to avoid duplication) • All ROCs agree with the concept, however few sites involved - so far • Requires commitment to upgrade quickly in production • As it is in production, might affect availability and reliability Process for retiring old versions of middleware services • Prototyped with Resource Brokers • Being implemented with gLite 3.1 services (replaced by gLite 3.2) • Big/medium sites able to follow new versions • Small ones slow, different issues, e.g. Obsolete H/W not running SL5 • Sites per Software version: 48% gLite 3.2, 52% gLite 3.1 List of gLite service versions supported by SA1 Operations https://twiki.cern.ch/twiki/bin/view/EGEE/SupportedServiceVersions SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  13. Distributed Operational Security Strong expertise in incident handling • Defined timelines for response and resolution • Security drills to educate and enforce procedures. Main source of incidents: failure to apply security patches • First alert in August 2009, EGEE completely patched after 2.5 months • First time a security patch was enforced at the sites • Escalated to EGEE-III PMB and WLCG Management Board • 60+ sites warned about suspension within 7 days, only 4 suspended • Second alert in November 2009, EGEE patched after 1 month • 30+ sites warned about suspension, none suspended Effective security patching monitoring • Security monitoring tools developed and deployed to prevent security incidents • Identify security weaknesses at sites to prevent incidents Extensive connections outside the project • Build links between ROC security contacts and local NRENs SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  14. Operational Security Support Grid Security Vulnerability Group • Help identify security weaknesses inorder to prevent incidents in deployed Grid middleware • 33 issues handled during the last year (60 since the start of EGEE-III) • Transition in progress to cover all the software distributed by EGI Joint Security Policy Group • 5 existing policies revised and 2 new polices: • VO Registration, VO Membership Management, Security Incident Response, Grid AUP, Site Registration • User-level Job Accounting, VO Portals • Security Policy Framework developed to support move to EGI International Grid Trust Federation • Bridged Research & Education federated ID and grid certificates • New guidelines for key management and the use of automated clients published SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  15. Global Grid User Support (GGUS) The EGEE helpdesk is a distributed system based on regional ticketing systems with central coordination and a central entry point for user support It is the only front-end exposed to end users • Developers & testers submit bugs directly to Savannah • GGUS tickets leading to Savannah bugs • Closed when Savannah bug reaches status “Ready for review”. • 44tickets ended in Savannah bugs since January 2010 • 28 of these for components under the responsibility of JRA1 Changes to support increased automation • Direct routing to sites (bypass 1st line support) • Reduction in tickets handled manually in the last year (2023 versus 3912) • ~80% of all 9419 tickets are now assigned directly • Additional automation options implemented when justified • Direct assignment to some VOs and GGUS Support Units • ALARM and TEAM tickets • Ticket escalation by submitter SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  16. Global Grid User Support Originating Community (Select VO from dropdown list) – Y2 None: tickets where the submitting user does not explicitly define the associated VO Atlas is the major user (31%), followed by LHCb(9%), CMS (3%) andbiomed (3%) Each VO has a different support model SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  17. Global Grid User Support User support tickets created in Y2 for each support unit type 1stline support (TPM) also solving 4% of the tickets 2nd line SpecialisedSoftware Support Units solving 8%of the tickets Most tickets solved at ROCs/sites, 87% SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  18. Grid Operator on duty Role of oversight and 1st level support for grid production infrastructure • Teams of operators on duty (COD) opening tickets on sites in response to grid monitoring alarms New model, based on the devolution to regions, fully implemented Summer ‘09 • First-line support done by each region, plus common layer for procedures, tools, escalation (the R-COD) Improvements: • Regional teams closer to the sites • Aware of different site deployment scenarios • More incentivised to prevent incidents • Shorter response time: 5.75 days (ticket assignment  closing) • Before regionalisation average response time 8 days • Distribution of responsibilities to ROCs  NGIs • Concern: Skills at ROC level will get partitioned when ROCs become NGIs SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  19. Grid Operator on duty Tickets opened in the last year Tickets opened per month NAGIOS • 2570 tickets opened by the R-COD in the last year with an average response time 5.75 days • Stable number of tickets opened by month, also after migration from SAM to Nagios • Move to new technology did not impact operations • Expect reduction in regional alarms as more sites start deploying Nagios • Sites see issues before they are reported to them by R-CODs SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  20. Grid Operations Tools • Aims • Improve reliability and availability of sites via operational tools • Prepare operational tools for use in an EGI/NGI structure • Achievements: • Better tools for site admins faster problem detection & solution • Operations dashboard with regional views, to raise alarms and GGUS tickets, used by R-COD • MyEGEEvisualization portal, collaboration with OSG • Availability/Reliability now calculated from regional Nagios results • Adoption and integration of mature open-source software • Site and Regional Monitoring via Nagios monitoring framework • Integration of operational tools via ActiveMQ messaging system SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  21. Grid Operations Tools Last 2 years: Now: SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  22. EGI: NGI prototyping Integration of NGIs into the infrastructure has started • Prototyped with 2 NGIs from 2 different multi-country ROCs • Poland from Central Europe • Greece from South East Europe • Centrally: Registration in managerial and operational structures • Locally: Tool deployment and adaption of processes (as required) Process now understood and documented Now being followed by many other NGIs • Issues: Tight coordination needed between tools and teams to enable the new NGI • Issues: With these two examples NGI staff overlapped with ROC staff with many years cumulated knowledge/experience • Most NGIs will need close support and training SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  23. Lessons learnt Regionalization is expensive due to extra layers • Move to national based infrastructures has ‘cheaper’ core Operation tools • Importance of agreeing on formats for information exchange between tools • Rely where possible on widely adopted public domain solutions (e.g. Nagios, ActiveMQ) • Need to have a more integrated operational tool development activity SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

  24. Summary Provided a stable and reliable infrastructure • While doubling (approx.) in capacity • Significant operational changes Need to improve operational tools • Home-grown tools & infrastructures costly to maintain and support Good transition to EGI, no major issues left • Full Implementation of ROC  NGI ongoing Global scientific collaborations relying on it • WLCG main user group, solid base for ESFRI engagement Collaboration was essential, important not to break this with regionalization/NGIs SA1 – Steven Newhouse- EGEE-III Final Review 23-24 June 2010

More Related